Automatic Mapping Clinical Notes to Medical - RMIT University
Automatic Mapping Clinical Notes to Medical - RMIT University
Automatic Mapping Clinical Notes to Medical - RMIT University
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
ous accounts of the effects of coherence relations<br />
on pronoun resolution have taken the view that<br />
the kind of relation shifts attention <strong>to</strong> different aspects<br />
of the event being described (Stevenson et<br />
al., 1994; Stevenson et al., 2000). If an event has,<br />
for example, a start-state and an end-state, then<br />
a narrative relation will shift attention <strong>to</strong>ward the<br />
start-state while a causal relation will shift attention<br />
<strong>to</strong>ward the end-state. Subsequent pronominal<br />
reference will therefore prefer referents associated<br />
with these respective states, as illustrated in (5)-<br />
(6). Based on this argumentation, the prediction<br />
would be that pronominal reference might favor<br />
one non-salient referent over another if it is associated<br />
with that part of the event <strong>to</strong> which attention<br />
has been shifted by the coherence relation.<br />
3 Experiment<br />
Before describing the experiment, I’ll review the<br />
primary and secondary questions which this experiment<br />
is designed <strong>to</strong> test. First, there is the<br />
question of what happens during pronoun resolution<br />
processes when there are competing nonsalient<br />
referents. Answers <strong>to</strong> this question should<br />
provide evidence <strong>to</strong>ward either a categorical or a<br />
gradient model of salience ranking. Furthermore,<br />
because investigating this question requires controlling<br />
for syntactic versus semantic prominence<br />
as well as coherence relation effects, two other<br />
secondary questions are also investigated. First,<br />
which is a more important fac<strong>to</strong>r in pronoun resolution:<br />
syntactic or semantic prominence? Second,<br />
what effect do coherence relations have on<br />
pronominal reference <strong>to</strong> non-salient entities?<br />
3.1 Design<br />
The research questions described above were investigated<br />
in this study using the well-known<br />
spray/load-constructions which exhibit the locative<br />
alternation (Levin, 1993) as shown in (7) and<br />
have synonymous alternative forms. 3<br />
(7) a. John sprayed some paint on a wall.<br />
b. John sprayed a wall with some paint.<br />
3 There is some difference of opinion on whether the two<br />
forms of spray/load-constructions are actually synonymous.<br />
One central point of contention is whether the <strong>to</strong>tality effects<br />
on the direct object (i.e., the judgment that the entity in direct<br />
object position is <strong>to</strong>tally used up in the event) are consistent<br />
across both forms. In the judgment of Rappaport and Levin<br />
(1988), the <strong>to</strong>tality effect applies only with the with-variant.<br />
In contrast, it is my judgment (Rose, 2005) and also that of<br />
Tenny (1994, see her data items (100) and (102)) that the effect<br />
applies across both forms.<br />
92<br />
According <strong>to</strong> prominence hierarchies in which<br />
the syntactic subject or the semantic agent is most<br />
prominent, then JOHN should consistently be regarded<br />
as the (most) salient referent while PAINT<br />
and WALL should be regarded as less or nonsalient<br />
referents in these sentences. Thus, subsequent<br />
pronominal reference with the third-person<br />
singular pronoun, it, allows a test of the three different<br />
questions outlined above.<br />
First, if a categorical approach <strong>to</strong> salience is sufficient,<br />
then there should be no overall preference<br />
for either PAINT or WALL. But if gradient salience<br />
is necessary for ranking, then it might be possible<br />
<strong>to</strong> observe a difference between the two.<br />
The nature of this difference, however, might<br />
be more complex depending on the way salience<br />
ranking is determined. If syntactic prominence is<br />
the only relevant fac<strong>to</strong>r, then preferences should<br />
consistently favor the object (i.e, PAINT in (7a),<br />
WALL in (7b)) according <strong>to</strong> the well-established<br />
syntactic prominence hierarchy in (2) above. But<br />
if semantic prominence is the only fac<strong>to</strong>r, then<br />
preferences should favor either the theme (PAINT)<br />
or the location (WALL) depending on how the semantic<br />
prominence hierarchy is ordered. One prediction<br />
might be based on proposed thematic hierarchies<br />
(cf., Larson (1988), Speas (1990)) which<br />
place theme above location. According <strong>to</strong> such a<br />
hierarchy, PAINT should be consistently preferred.<br />
This is what I observed in Rose (2005).<br />
Other differences may result from the kind of<br />
coherence relation used. However, for spray/loadconstructions,<br />
this is a little difficult <strong>to</strong> predict.<br />
The two non-salient entities are both arguably a<br />
part of the end-state of the event—that is, <strong>to</strong>gether,<br />
they are the product of the agent’s work. Thus, any<br />
motivation <strong>to</strong> distinguish between the two with respect<br />
<strong>to</strong> the coherence relation must come from<br />
some other feature of the event or its participants.<br />
I will address the possibility in the discussion section<br />
below.<br />
3.2 Method<br />
3.2.1 Participants<br />
The participants in this experiment included 36<br />
undergraduate students at Morehead State <strong>University</strong><br />
in Kentucky. Students were recruited through<br />
fliers and classroom announcements and received<br />
five dollars for their participation.